Layered picture and sound
Place clips on tracks, trim their ranges, split shots, and change source in-points. Keep picture edits and audio placements in the same sequence.
TOOLS DEVELOPMENT / NODE-BASED CREATION
Dreamkeeper
A node-based workspace
for images, video, and 3D.
Dreamkeeper combines compositing, painting, generative AI, 3D scene assembly, animation, editing, and audio in a native desktop application. Typed graph connections carry images, masks, text, clips, geometry, and scene data between tools.
Explore video editing ↓01 / VIDEO EDITING
Cut, layer, and finish video
on a connected timeline.
Arrange footage on layered picture and audio tracks. Trim and split clips, set source in-points, and use adjustment clips for shared treatments. Each shot can connect to Dreamkeeper’s compositing, grading, animation, and audio tools.
Graph-based compositing and effects
Connect shots to grades, masks, composites, and animated titles. Use adjustment clips to apply a shared treatment across the edit.
Combine imported or generated footage, add procedural titles and VST3-processed audio, then publish a versioned movie. Integrated playback, presentation, and device acceptance remain in progress.
02 / LIBRARY
Build a library of reusable tools.
The Library organizes Node Presets, Brush Presets, and Skills in a dedicated browser. Save useful graph setups, bring them into another project, and keep recurring creative workflows close at hand.
Reuse connected workflows
Store graph fragments and reusable groups as presets, then insert them into the working graph. A saved tool can retain its internal connections and exposed group controls.
Keep creative settings together
Browse and apply brush presets, or reuse Skills that package instructions and inputs for supported AI workflows. These sit alongside node presets as distinct Library categories.
03 / GROUP INTERFACE EDITOR
Give a node group its own interface.
The Group Interface Editor turns a connected graph into a focused tool. Promote internal parameters, bind controls to their targets, and define public inputs and outputs through the Parameters, Inputs, and Outputs tabs.
Expose the controls that matter
Set labels, defaults, ranges, and descriptions. Arrange controls into nested sections and tabs, choose horizontal or vertical layouts, and add visibility or enablement rules.
Share the design, vary each instance
Preview the draft before applying it as an undoable edit. Linked groups share the Interface design while keeping independent values, animation, and image connections. Save the group to the Library for reuse.
04 / BATCH & CONDITIONAL WORKFLOWS
Apply an editable graph to a whole list.
For Each runs its body once per item and collects results in the original order. Lists can carry images, text, geometry, and other graph values. Inspect a selected item inside the loop while retaining the same reusable processing graph.
Choose which processing runs
Numeric Compare and Boolean Logic drive Switch Any. For example, resize only images above a chosen dimension while passing smaller images through unchanged.
Package the operation as a tool
Use list filtering, text formatting, expressions, and group interfaces to expose useful batch settings. Save the setup to the Library. For Each is an ordered list map with sequential execution.
05 / AI CREATION
Generate, revise, and connect AI results.
Build image and text workflows from prompts, reference images, and reusable Skills. Image Generation connects OpenAI or Google providers to the graph; LLM Inference and LLM Model nodes supply text and shared model settings for downstream tasks.
Work from a saved result
Select an image from history, describe a change, and add references or painted annotations. Branch from earlier results and retain revision ancestry. Saved request recipes preserve the provider, inputs, and settings used for a retry.
Turn animation into generation guidance
Capture animated text, masks, patterns, or rendered imagery as control footage for a supported video model. Source video and reference images provide other conditioning paths. Model guidance is approximate; exact shapes can remain in the compositing graph.
Generation uses explicit actions and saved caches. Video, depth, pose, upscale, and outpaint operations depend on their selected models or workers. New LLM conversation functionality remains under acceptance.
06 / ANIMATION & MOTION GRAPHICS
Animate parameters, characters, and whole shots.
Use the Graph Editor to keyframe supported parameters, shape curves and tangents, and combine animation layers. Expressions and sampled animation drive procedural relationships without duplicating every operation by hand.
Animate text and sprites
Text Layout and Image Elements feed element fields and transforms before Render Elements rasterizes them. Per-character delays, spacing, opacity, and blur build titles from connected operations.
Fit motion to the edit
Reusable title groups can retime their reference animation to a clip’s length. Keep animation curves editable, expose timing controls on the group, and add 2D or 3D camera shake where needed.
07 / 3D MODELLING & RENDERING
Build editable scenes and return them to the composite.
Assemble imported geometry or native primitives, edit geometry in the viewport, and organize cameras, transforms, lighting, and materials through the graph. Character, mannequin, and Pose to Rig tools support posing workflows; image-to-pose supplies a typed pose and a visual pose map through a configured backend.
Render more than the beauty image
Combine procedural textures and materials, then render a scene into the image workflow. The render-pass framework defines beauty, depth, normals, ambient occlusion, object IDs, and albedo for supported paths.
Load, reconstruct, and render
Import splats, transform or crop them, render camera views, and export the result. Image-to-Gaussian and image-set camera solving/training add reconstruction paths with their required services or workers.
08 / COMPOSITING, PAINT & MATTES
Paint corrections, isolate subjects, and finish the image.
Combine merge, transform, crop, resize, blur, and glow operations with layered painting. Pressure-driven brushes, textured tips, blend modes, clone/heal/smudge tools, and selections support direct image work alongside procedural processing.
Separate selection from edge refinement
Use shape masks and mask combinations, or connect an external SAM segmentation backend. Learned matting and refinement tools handle soft edges, feathering, grow/shrink, cleanup, despill, defringe, and light wrap. Video masking has its own configured backend path.
Grade the composite
Color Curves and Color Wheels provide curve-based controls, primaries, log, and HDR adjustments. Film treatments include grain, response, halation, split toning, and vignette; Color Match supports reference-based balancing.
09 / STORYBOARDS & REFERENCE SHEETS
Develop panels, timing, and reference layouts.
Storyboard Builder combines artwork, panel descriptions, and hold durations. Draw with onion-skin context, edit panel timing, and choose sheet layouts, captions, and title styling. Storyboard to Clip turns the timed panels into an animatic; Slicer extracts individual panels and Video to Storyboard samples existing footage.
Work between frames and sheets
Use the drawing and onion-skin view while developing panels, then inspect the board as a sheet or a playable clip.
Arrange labeled visual references
Build hero/detail or grid layouts with editable image framing, crop focus, labels, and title placement. Save layout presets and feed the finished sheet into another image or generation workflow.
10 / AUDIO & VST3
Build a soundtrack inside the project.
Connect audio sources through effects, mixes, sends, buses, and returns. Audio FX hosts VST3 effects in a separate Windows worker, with native plugin editors embedded in Properties or opened separately. The current processing path uses 48 kHz stereo.
Keep treatments editable
Group effects into reusable racks and expose useful controls. Supported live audition paths provide automation reading, override/touch behavior, and recording of supported parameter gestures.
Retain sound and its recipe
Freeze a treatment into rendered audio while retaining its editable recipe, unfreeze it to revise, and audition alternatives. Publish processed audio with the movie.
Live playback, plugin compatibility, and device acceptance have defined limits. MIDI/instrument delivery remains deferred.
11 / HISTORY, ASSETS & PUBLISHING
Keep creative revisions and publish named deliveries.
Image History retains generated results and their revision ancestry for later selection or editing. Output publishing creates versioned deliveries, while Asset nodes bring published work into downstream graphs. These give intermediate experiments and finished outputs distinct roles in a project.
Return to the decision behind an image
Inspect saved variants, select a revision source, and retain the generation recipe. Reopening stored results does not automatically submit another model request.
Reuse published work
Publish images or movies with output settings and presets, inspect versions in the Publish Browser, and reference assets from another workflow. Save the editable project alongside the delivery process.
12 / IN DEVELOPMENT
Agent-built graphs and new ways to interact.
These integrations are under development or retain open acceptance checks. They illustrate the project’s direction with editable native work and explicit artist control.
Create editable work from a task
Connect a model and Skill to an enabled Agent Group, then explicitly start a run. The agent can construct or revise supported native nodes, inspect rendered evidence, and retain checkpoints. The image-modelling preset works from reference images; live quality and release acceptance remain in progress.
Extend scenes and capture interaction
Environment extension continues approved geometry as editable scene objects. Phone integrations explore tracked camera control with returned scene video; VR work explores scene viewing and hand interaction. Physical-device and release acceptance remain separate.
13 / TECHNICAL FOUNDATION
Native UI. Vulkan image processing.
Dreamkeeper uses C++ and Qt for its desktop interface, Vulkan compute for core image operations, OpenImageIO for image I/O, and FFmpeg-based media processing. Supported processing and viewer paths retain GPU-backed results; export, external services, and fallback consumers have explicit transfer boundaries.
Generation starts through explicit actions. Cache mode restores saved results, while “Generate up to here” evaluates supported upstream generation dependencies for an Output and reuses matching completed steps. This lets a project retain expensive results while its surrounding graph remains editable.
Inspect the image and its dependencies
Pan and zoom the Viewer, inspect channels, and adjust display exposure/gamma. Follow intermediate graph outputs, compare results, inspect metadata, or profile a branch. Resolution and pixel precision can inherit from the graph or primary input, with explicit overrides at supported boundaries.
14 / DEVELOPMENT SCOPE
Current scope and dependencies.
Which features are still undergoing acceptance?
Timeline playback/presentation and VST3 workflows retain open acceptance checks. Agent Groups can create editable native graphs through model-driven tools, but live quality and release acceptance remain in progress. Phone camera/scene streaming and VR interaction are experimental integrations. MIDI/instrument delivery is deferred.
Do I need AI to use it?
Compositing, painting, grading, animation, and scene tools have independent workflows. Generative nodes require the selected provider’s credentials or a configured local worker; SAM, depth, pose, and reconstruction tools have their own backend/model dependencies.
What is saved and reusable?
Projects retain authored graphs and settings. Image History preserves generated results; the Library stores reusable graph fragments, brushes, and Skills. Property presets transfer node settings. Output publishing produces versioned assets for downstream use.